Holo3.1: a fast local computer-use agent
Original title:Holo3.1: Fast & Local Computer Use Agents
AI summary
Holo3.1 is a family of computer-use agents built on the Qwen model family, designed to improve robustness in desktop, web, and mobile environments. The new models come in four sizes — 0.8B, 4B, 9B, and 35B-A3B — and release quantized checkpoints for the first time, including FP8, Q4 GGUF, and NVFP4, to optimize local inference. On the AndroidWorld benchmark, the 35B-A3B model's score rose from 67% to 79.3%. On DGX Spark, NVFP4 quantization delivers a 1.74x token throughput improvement over BF16 and cuts average step time from 6.8 seconds to 3.3 seconds. The models support function-calling protocols and can be deployed in third-party agent frameworks.
Key Takeaways
- 1 is a family of computer-use agents built on the Qwen model family,designed to improve robustness…
- The new models come in four sizes — 0.
- 8B,4B,9B,and 35B-A3B — and release quantized checkpoints for the first time,including FP8,Q4…
- On the AndroidWorld benchmark,the 35B-A3B model's score rose from 67% to 79.
- On DGX Spark,NVFP4 quantization delivers a 1.
Quick Facts
- Published
- Updated
- Source
- Hugging Face 社区博客(混合发现)
- Category
- Model Releases
- AI score
- 73
Holo3.1 把计算机使用代理从桌面扩展到了移动端,还首次放出了量化版,让本地运行真正快了起来。想做 GUI 自动化的开发者可以立刻跑起来了。
Related entities
Source & copyright
This article is based on curated titles and AI summaries from the AIHOT public API, with attribution and links back. Original copyright remains with the original publishers and authors; this site does not mirror full texts.
How to Cite
When citing this article, keep the title, date, and permalink, and credit Orange News as the source.
Orange News: Holo3.1: a fast local computer-use agent. June 2, 2026. https://new.nat6.net/en/news/cmpwqvwfa05cqslsncgvlgd9q/